Gemini 4 Argon Reaches Google AI Mode for AI Pro and Ultra Subscribers

Google said on 13 October 2026 that Gemini 4 Argon, its newest frontier model, is now available in AI Mode in Search for Google AI Pro and Ultra subscribers, and that the rollout is global. Nothing announced that day changes the default model for free AI Mode users. For SEO and GEO teams, the practical change is that a different model now writes some AI Mode answers for paying users, and early reports say those answers carry links to sources.

The announcement came in a post on Google's The Keyword blog written by Koray Kavukcuoglu, SVP of Google DeepMind and Chief AI Architect at Google. The same post opened Argon to developers through the Gemini API, to enterprises, to Google AI Ultra subscribers in Google Antigravity, and to Pro and Ultra subscribers in the Gemini app. This article focuses on the Search side, because that is the part that touches organic visibility, and keeps the developer detail short.

What Google announced on 13 October 2026

Google described Gemini 4 Argon as "globally available today" across several surfaces at once. For everyday users, the post says to "try it starting today in the Gemini app and AI Mode in Search" with a Google AI Ultra or Pro subscription. Google did not attach a country list or a language list to that statement.

Robby Stein, VP of Product for Google Search, announced the Search rollout on X on the same day, writing that "Gemini 4 Argon is now rolling out globally in Search!" He said Argon "brings big improvements to tackling nuanced, multi-step tasks in AI Mode" and is better at understanding intent and following instructions in back-and-forth conversations. According to trade-press reporting of that post, subscribers switch to Argon through the "+" icon in AI Mode's "Ask anything" bar, the same model menu that received Gemini 3.7 Flash on 14 August and Gemini 3.8 Flash on 2 September.

Google first introduced Argon about two weeks before this release, with access limited to a set of trusted cyber defenders through its Fairwind Program. The 13 October post moves it to paying consumers, developers and enterprises in one step, and says Google is "beginning expansion to hundreds of additional Fairwind Program partners."

Who can use Gemini 4 Argon, surface by surface

The table below lists each access route Google named in the 13 October 2026 post, with the conditions Google attached to it.

One caution on the first row. "Rolling out" means staged. Trade-press reporting on launch day noted that Argon had not yet appeared in the model menu of at least one Pro account as of 12:17 p.m. US Eastern time. If it is missing from your own account, that matches the rollout pattern and is not a sign that your region is excluded.

Links in AI Mode answers: why this matters for citations

Search Engine Roundtable reported on 13 October 2026 that AI Mode responses generated with Argon include links, and contrasted this with the earlier Gemini 3.8 Flash release in AI Mode. That is the detail SEO teams should watch most closely. An AI Mode answer that cites sources can send a click to a publisher or a brand. An answer that cites nothing sends nobody anywhere, however accurate it is.

The limits of what is known matter here. Google's post does not say whether Argon changes which sources AI Mode cites, how many links each answer shows, or how link placement compares with the default model. It also does not say whether Argon handles every AI Mode query a subscriber makes once selected, or only some of them. PPC Land noted that nothing has been published to measure Argon's effect on citations. The honest reading today is: links are present, and their volume and selection are unknown.

What "multi-step tasks" means for how pages get used

Google and Stein both frame Argon around nuanced, multi-step tasks and follow-up conversations. In AI Mode, a multi-step query is the kind where a user asks one thing, refines it, adds a constraint, then asks for a comparison. Each turn can pull in new sources. A page that answers only the first, broad question may be cited once and then dropped as the conversation narrows.

Analysis, not a Google statement: pages that cover the follow-up layers of a topic (criteria, trade-offs, exceptions, local conditions, prices with dates) give a model more reasons to keep citing them as a conversation goes deeper. That is a content-depth argument, and it applies with or without Argon, but a model marketed on multi-step reasoning makes it more relevant.

The writing style Google describes

Google's post says Argon "answers questions more directly with less hedging, doesn't repeat itself more than necessary, and skips clichés throughout the responses," and that it follows style instructions better than any previous Gemini model. For publishers, a more direct answer can cut both ways. A tighter answer leaves less room for padding, so the facts it lifts need to be specific and attributable. A page that states a figure with its date and its source is easier to quote than a page that circles the point.

How Argon fits the AI Mode model menu

AI Mode has run on several Gemini models since late 2025, and the pattern helps explain where Argon sits. According to Search Engine Journal's running coverage, Gemini 3 Pro became an option in AI Mode in November 2025 and stayed an option. Gemini 3 Flash became the default in December, and Gemini 3.5 Flash became the global default at Google I/O in May 2026. Gemini 3.7 Flash and 3.8 Flash joined the menu as subscriber options in August and September 2026.

Argon arrives the way Gemini 3 Pro did: as a menu option, not as the new default. Neither Google's post nor Stein's X post says whether that will change. The post ends by saying Google is "looking forward to bringing you the rest of the Gemini 4 family," without naming which of those models will reach AI Mode or when.

The same reporting noted that Google's help page for the AI Mode model menu still listed only "Pro" and "Fast" options on launch day, describing "Pro" as Gemini 3 Pro. Documentation lagging the product is common, so treat the model menu itself as the source of truth for what a given account can select.

What Google said about performance

Most of the performance claims in the 13 October 2026 post are Google's own measurements and cannot yet be checked from outside. The post says Argon ranks first on the Vals Index, which Google describes as measuring model performance in finance, coding, legal and tax work. Inside Google, the post says, Argon in Antigravity helped teams raise agentic code submissions by more than 35% while cutting code rollbacks by nearly a third. No baseline or sample size is given for either figure.

Some examples are more checkable because the code is public. Google says Argon helped the Chrome team separate core webpage handling from a roughly 6 million-line legacy folder inside Chromium's 40 million-line codebase, and helped the team behind the open-source Agent Development Kit cut in-memory session preparation time by 40% in long sessions and speed up cold imports by 21%. Those ADK changes are, per Google, live in the google/adk-python repository.

Developer and API details, briefly

For teams building tools on the Gemini API, three facts from the post matter. First, the introductory price is $2 per million input tokens and $10 per million output tokens, with cached input tokens at 95% off. Second, a footnote in the post says that after the introductory period expires on 30 November 2026, the price becomes $4 per million input tokens and $20 per million output tokens; it applies to the Argon model intended for public access. Third, Google cites a 1M unified input-output token limit via the Gemini API and points developers to documentation on a "new continuation protocol." Access runs through the paid tier of the Gemini API in AI Studio and Android Studio.

For SEO teams that run content analysis or brief generation on the API, the 30 November date is the one to put in a calendar. A budget built on $2 and $10 doubles in December.

What this means for Thai marketers

Google's post says Argon is rolling out "globally" in AI Mode for Pro and Ultra subscribers. It does not list countries, and it does not list languages. Trade-press coverage pointed out that neither Google's post nor Stein's X post states a language, and that the earlier Gemini 3.7 Flash rollout in AI Mode was English only. So Google has not said whether Argon answers Thai-language queries in AI Mode, and this article does not assume that it does.

One practical point follows. The audience that sees Argon answers is narrow: only people in markets where the rollout lands, who pay for Google AI Pro or Ultra, and who switch the model on. For most Thai consumers searching today, nothing about their AI Mode experience has changed.

What to check in your own work this month:

• Run your priority English queries in AI Mode on a Pro or Ultra account with Argon selected, then again with the default model, and record which domains each answer links to.
• Repeat the same queries as multi-turn conversations (broad question, then two narrowing follow-ups) and note where your pages drop out.
• Check that your pages state facts with dates, figures and named sources in plain sentences, since Google describes Argon as more direct and less hedged.
• If you test in Thai, record whether the Argon option appears at all before drawing conclusions about Thai answers.
• If you run tools on the Gemini API, note the 30 November 2026 price change.

A structured AI SEO programme turns this kind of side-by-side testing into a repeatable process. If citations in AI answers are already a reporting line for your brand, a generative engine optimization (GEO) audit is the place to start, and the same checks feed into standard SEO in Thailand work.

What Google did not say

Several gaps are worth stating plainly, because they decide how much weight to give this launch:

• Google did not list the countries or languages included in the "global" AI Mode rollout.
• Google did not say whether Argon will become the default AI Mode model for free users.
• Google did not say whether Argon changes which sources AI Mode cites or how many links it shows.
• Google did not publish Search-specific figures, such as click-through from Argon answers.
• Google did not name the other Gemini 4 models or their dates.

FAQ: Gemini 4 Argon in AI Mode

Is Gemini 4 Argon available in AI Mode in Thailand?

Google has not said. Its 13 October 2026 post says Argon is rolling out "globally" in AI Mode for Google AI Pro and Ultra subscribers but does not list countries or languages, so the only reliable check is whether the Argon option appears in the "+" model menu on a subscribed account.

Do free AI Mode users get Gemini 4 Argon?

No, not from this announcement. Google limited Argon in AI Mode and the Gemini app to Google AI Pro and Ultra subscribers, and nothing announced on 13 October 2026 changes the default model for free users.

Do Argon answers in AI Mode include links to websites?

Yes, according to Search Engine Roundtable's 13 October 2026 report, which said Argon responses in AI Mode include links, unlike the earlier Gemini 3.8 Flash release. Google has not published how many links appear or how sources are chosen.

Do I need to change my SEO because of Argon?

No emergency change is needed, because only opted-in paying users see Argon answers. It is a good reason to test priority queries with Argon and the default model side by side and to check how your pages hold up in multi-step follow-up conversations.

How much does Gemini 4 Argon cost on the API?

Google set an introductory price of $2 per million input tokens and $10 per million output tokens, with cached input 95% off. After the introductory period ends on 30 November 2026, the price rises to $4 and $20 per million tokens.

The short version

Gemini 4 Argon is now an opt-in AI Mode model for Google's paying subscribers, announced as a global rollout on 13 October 2026, and early reports say its answers link out to sources. It is not the default, Google has not named the languages covered, and Google has published no data on its citation behaviour. Test it on the queries that matter to you, keep notes on which domains it cites, and revisit when Google says more about the rest of the Gemini 4 family. If you want help building that testing into your AI search reporting, Relevant Audience can walk you through it.